Why Do Shareholder Votes Matter? 1

Transcription

1 Why Do Shareholder Votes Matter? 1 Laurent Bach & Daniel Metzger December 2014 Abstract We show that majority-supported shareholder proposals create value not necessarily because their content has intrinsic value, but because they increase pressure on the board of directors. We document that shareholder organizations (CII), proxy advisory firms (ISS), and management often disagree about voting results because they apply different majority requirements in 60% of the cases. This allows us to identify each of those key players reactions to what they consider an approved proposal. As soon as CII considers a proposal has passed, there is higher pressure on noncomplying boards, as votes against directors and inefficient CEO turnovers increase very sharply. This is overall well-perceived by investors as stock prices also go up. Sanctions taken by ISS remain mild against otherwise well-perceived directors and do not cause inefficient CEO turnovers, but they do not improve firm value either. Interestingly, those effects of passage according to CII and ISS arise only when a proposal is not eventually implemented. 1 Stockholm School of Economics & Swedish House of Finance, Drottninggatan 98, Stockholm, Sweden. laurent.bach@hhs.se; daniel.metzger@hhs.se. We thank Marco Becht, Vicente Cuñat, Daniel Ferreira, Ernst Maug, Per Strömberg, Paolo Volpin, as well as seminar participants at HEC Paris, FU Amsterdam, Nova/Catolica Lisbon, WHU Vallendar, IFN Stockholm, Frankfurt U, Stockholm U, 2012 Gerzensee Evening Sessions, AFFI-EUROFIDAI 2013 conference, 15th Workshop on Corporate Governance in Oslo, and at the Swedish House of Finance for very helpful comments. This paper was initially circulated with the title The Dark Side of Shareholder Activism: Evidence From CEO Turnovers. 1

2 1 Introduction Voting in general assemblies of shareholders has long been considered a poor way of disciplining managers. High-profile commentators of corporate life such as Bebchuk (2007) go as far as considering the shareholder franchise as just a myth. A large part of the literature on shareholder activism instead focuses on activism led by hedge funds seeking major changes in management and corporate strategy (Gillan and Starks, 2007). We believe this relative disdain for low-cost voting activism needs to be reevaluated in the light of recent developments in US corporate governance. Since 2001, the number of S&P 1500 firms in which a shareholder proposal wins a majority of votes cast against the advice of management has more than doubled and is now close to a hundred every year, while hedge fund activists generally target about fifteen of those big firms every year. 2 More strikingly, our own numbers show that the rate of implementation by management of majority-supported governance proposals has increased from around 20% before 2001 to close to 70% in Recent academic research (Cuñat, Giné, and Guadalupe, 2012, 2013) has also documented large and positive valuation impacts of majority support for shareholder proposals. Those effects have been primarily ascribed to the impact of the proposal content actually being implemented. In this paper we propose, and find empirical support for, an alternative mechanism for how shareholder support for a proposal affects firm value: majority support for shareholder proposals, if not significantly addressed by the board, undermines the confidence that passive investors had previously put in management and in so doing reduces its entrenchment. We are able to identify this surge in defiance among passive shareholders following a majority vote because a number of institutions dedicated to those investors (e.g., shareholder organizations, proxy advisory firms, labor unions, shareholder online forums, and blogs) specifically keep track of proposals that receive the support of more than 50% of the sum of votes cast for and against it. Some of those entities take this monitoring to the next stage and track the implementation of majority-approved proposals in the year following the meeting on behalf of their members and clients. Based on this information, boards decisions following majority votes are benchmarked against best practices of corporate governance, which may significantly affect directors reputation. For instance, the Council of Institutional Investors (CII), which represents a large portion of American pension funds, recommends that boards implement a proposal as soon as more than 50% 2 See Figure 1. We thank Fos (2013) for providing his data. 2

3 of votes cast for and against it and it keeps a list of non-complying firms available to its members. Boards singled out by CII may then be the target of coordinated vote-no campaigns. The main proxy advisory firm, Institutional Shareholder Services (ISS), has a policy of recommending a vote against incumbent directors if they fail to implement a proposal that has received the support of a majority of outstanding shares. Using a large sample of shareholder proposals voted upon in the US between 1997 and 2011, we provide evidence that majority-supported proposals have substantial effects on the level of entrenchment of directors. If support for a proposal barely passes 50% of votes cast for and against (the CII threshold) or 50% of outstanding shares (the ISS threshold), i.e., votes cast for and against but also abstentions, non-votes and shares absent at the meeting, and yet management does not implement the proposal, then votes against directors in the next election may increase by up to 120% with CII and 80% with ISS. Interestingly, not all directors are affected equally: the sanction imposed by ISS following lack of board responsiveness disproportionately affects the level of defiance against the least popular nominee, while the vote-no campaigns triggered by CII primarily affect the amount of defiance votes received by the most popular nominee. We conjecture that this is because the ISS board responsiveness policy can be relaxed on an individual basis if ISS detects that some nominees are particularly talented, while CII policies target boards as a whole and not individual directors. Since director elections are very rarely contested in the US (Becker and Subramanian, 2013), the strong impact of a majority-approved shareholder proposal on future director elections may bear very little consequences for the board members and, ultimately, for firm value. On the other hand, an abnormal amount of votes withheld from a director may be enough to damage her reputation and lead her to take corrective action (Grundfest, 1993). Our data backs the latter claim as we find strong evidence of effects of majority support as defined by CII (i.e., more than 50% of for and against votes) beyond director elections. Crossing this threshold of support triggers a positive abnormal return of about 1% on the day of the meeting. It also makes it much more likely that CEOs subsequently leave the firm to the detriment of firm value, i.e., with a significant stock price decline on the day of announcement of the CEO departure. Our interpretation of those seemingly contradictory effects is as follows. After a proposal wins a majority of votes for and against the proposal, and the board does not respond, CII members essentially commit to penalizing all 3

4 directors irrespective of their individual performance. This commitment is credible because CII members are dispersed and for that reason will not be able to agree on a divergence from their initial policy. An indirect effect of those penalties may be that CEOs themselves have to leave despite having a good track record. However, this commitment taken by CII to defy current directors and officers is also a strong disciplining tool: in order to counteract such a strong defiance and save their reputation, management has to take extra actions genuinely benefiting shareholder value. Interestingly, our data shows that sanctions of directors coming from ISS do not have any effect on stock returns or CEO turnover. This is because those sanctions can be cancelled by having an otherwise good individual track record, so that management may often feel too safe to take additional value-enhancing steps. An alternative hypothesis for those findings is that majority approval of a shareholder proposal according to CII triggers its implementation by management and, hence, improves formal governance. This view is taken by Cuñat, Giné, and Guadalupe (2012, 2013, and 2014), Cheng, Hong and Shue (2014), Ertimur, Ferri and Oesch (2014), and Popadak (2013). For instance, Cuñat, Giné, and Guadalupe (2012) identify a significant and positive stock market reaction to shareholder proposals closely reaching the threshold of approval according to the corporate bylaws (henceforth the Management threshold). 3 They also find that such events are followed by a general erosion of takeover defenses. From this they conclude that it is this expected formal governance improvement which is positively perceived by the market. If the majority concepts used by CII and management were identical, this corporate governance explanation to our results would be impossible to distinguish from our proposed director defiance explanation. However, collecting data on the voting rule in the corporate charter, we are the first to document that in about 60% of the cases, the Management threshold of approval is higher than the CII threshold because abstentions, broker non-votes or shares absent to the meeting count as votes against the proposal according to the bylaws while those voting choices are not tallied by CII. As a result, in about 5% of the cases a proposal has passed according to CII but not according to management. 3 This Management threshold is set by the voting rule (e.g., treatment of abstentions or non-votes) as defined in the corporate charter and may vary across firms (and even firm-years). 4

5 Those disagreements are very informative: we show that as long as a proposal has not received majority approval according to the bylaws, it is only very rarely implemented, even when the proposal has passed according to CII; on the other hand, as soon as a proposal has reached the majority threshold defined by CII, and even if it has not yet been approved according to the firm s bylaws, there is a positive impact on stock returns on the day of the meeting, defiance votes in future director elections, and higher CEO turnover. This means that the board feels sufficiently strongly against the proposal content that it is ready to risk being punished by investors in order not to implement the proposal. Moreover, our data shows that most of the effects of majority approval according to CII or ISS have some bite only when the firm decides not to implement the proposal, which is further evidence that it is the punishment of unresponsive directors by third parties which is valuable rather than the expectation that some of the boards will implement the proposal. Because we investigate how a non-binding proposal can have real consequences, we see our contribution as the first empirical counterpart to the theory of shareholder proposals articulated by Levit and Malenko (2011). Their model concludes that voting support for such proposals does generally not force management to implement them unless there is a shareholder activist who can seize the voting outcome as an opportunity to replace current management. Interpreted broadly, our results echo this prediction, as voting results seem to matter for value only because some intermediary (a shareholder organization or a proxy advisor) threatens board members with their removal upon learning the result of the vote. However, one major point of divergence is that we find that most of management s reaction to the vote takes place discontinuously around a salient majority threshold defined by the bylaws. In other words, managers use rule-of-thumbs instead of deep strategic thinking when choosing what stance to take after the vote, either due to reputational concerns and/or optimization costs. Our work also fits into the literature on corporate voting rules and their impact on board entrenchment. Maug and Rydqvist (2009) show that shareholders tend to vote more aggressively when the corporate charter imposes a super-majority threshold, in such a way that the super-majority requirement tends to be a poor way of entrenching management. We go further and find that even the interpretation of a given voting result is not solely left to the corporate charter, so that management cannot rely on internal laws of the firm to dismiss the outcome of a vote. 5

6 While a large part of the empirical literature on activism has dismissed the importance of voting as a monitoring tool (Black, 1998; Karpoff, 2001; Romano, 2001; Gillan and Starks, 2007), our paper takes place in a burgeoning literature documenting the effectiveness of voting activism, surveyed in Yermack (2010) and Ferri (2012), for instance. Ertimur, Ferri and Stubben (2010) document the negative impact of the lack of board responsiveness to approved shareholder proposals on directors reputations. Cai, Garner and Walkling (2009) find strong effects of withheld votes in director elections on CEO turnover, while Del Guercio, Seery and Woidtke (2008) show that voteno campaigns, which are often triggered by the absence of a response to majority-approved shareholder proposals, increase the likelihood of a forced CEO turnover. Those papers precisely document how elections can serve to punish unresponsive directors and CEOs, but none of them have delivered quasi-experimental evidence of this mechanism. More importantly, our research uncovers a key trade-off in voting activism: punishments of directors and CEOs through director elections may be ex-post inefficient as shareholders voting guidelines are often very crude, but the commitment to exert this punishment is ex-ante beneficial because it erodes the entrenchment of management and in so doing delivers extra pressure to act for the benefit of shareholders. Finally, this paper provides essential input for current policy debates on the role of proxy advisors. There is currently a move to regulate these firms as they are accused to have too much weight in shareholder votes given the poor quality of the recommendations they issue (US House, 2013). While the literature has estimated very strong correlations between voting recommendations from proxy advisors and voting results for director elections (Cai, Garner and Walkling, 2009; Choi, Fisch and Kahan, 2010; Li, 2013), it is hard to see those as causal since many shareholders vote like proxy advisors just because they share the same information set. We show that voting results on shareholder proposals provide an exogenous source of variation in ISS recommendations, and it allows us to estimate causally that a recommendation to vote against a director translates into an additional 25 percent of votes being withheld, arguably a very sizeable effect testifying to the current power of proxy advisory firms. On the other hand, we also provide evidence that ISS recommendations do take into account individual-specific information to the benefit of current board members: good directors are exempted from ISS sanctions when the board as a whole has failed to respond to a past shareholder proposal. This means that the popular criticism according to which proxy advisors use one-size-fits-all voting guidelines is not well-grounded. Moreover, our evidence on the effect of CII policies shows that if ever proxy advisors voting guidelines became 6

7 too rigid, this would also provide a valuable way of disciplining management and therefore should not be prevented. The remainder of the paper is as follows. Section 2 describes the institutional details of voting on shareholder proposals in the US. Section 3 develops our hypotheses. Section 4 presents the data and the variables used in the analysis. Section 5 discusses the empirical strategy as well as its internal and external validity. Section 6 presents the results. Section 7 concludes. 2 Institutional Background Our empirical analysis and the interpretation of our results regarding the impact of shareholder votes rely heavily on a few specific features of shareholder voting in the US. Therefore, a detailed explanation of the voting process and its repercussions is in order. This will allow us to derive testable hypotheses on the impact of shareholder support for shareholder-sponsored proposals. 2.1 What do Shareholders Vote on? On the occasion of general assemblies, shareholders can be asked to vote on many different matters. They may elect directors or vote on specific proposals, sponsored either by the management or by a shareholder. Our study is centered on shareholder-sponsored governance proposals. Those obey to SEC rule 14a-8 and are different in several dimensions from management-sponsored proposals that have been studied in related papers (e.g., Popadak, 2013). The main difference is that shareholder-sponsored proposals are usually not binding the management. This means that even if the approval threshold set by the corporate charter has been passed, the board of directors has discretion over whether or not to implement the proposal. This matters because in the overwhelming majority of cases, management recommends shareholders to vote against shareholder proposals. The paradox that shareholder proposals are officially nonbinding and yet have been shown to carry real effects (Cuñat et al., 2012, 2013) is what motivates our focus on this type of governance proposal. 2.2 How are Votes counted? Even though shareholder proposals can never bind the board s final decision, corporate charters specify a threshold that the number of votes for the proposal needs to reach before it is considered passed and, as such, worthy of consideration by the board. 7

8 Faced with an incoming voting decision, shareholders have five different choices: 1) send a proxy marked For (F), 2) send a proxy marked Against (A), 3) send a proxy marked Abstain (AB), 4) let the broker send the proxy on their behalf without any indication (BNV), or 5) not send a proxy either directly or indirectly (NP). These votes are then aggregated according to a voting rule to decide whether a proposal has been approved or not. A voting rule essentially determines whether certain voting options (AB, BNV, and NP) are not counted or de-facto counted as votes against a proposal. Given the different voting options, there can be four different voting rules (or approval thresholds) 4 : Threshold 1: F/(F+A) > 50% Threshold 2: F/(F+A+AB) > 50% Threshold 3: F/(F+A+AB+BNV) > 50% Threshold 4: F/(F+A+AB+BNV+NP) > 50% We collect data on the voting rule that is used by each firm from proxy statements before each meeting. The corresponding descriptive statistics are presented in Section 3 and show a very significant dispersion of voting methods across firms. This is intriguing since state corporate law sets a default rule regarding when proposals are considered to be passed. However, firms are always set free to opt out of the state default and choose a different approval threshold and this is an option that they seem to exercise very often. Indeed, while the corporate charter typically specifies that votes for should represent more than 50% of votes cast, there are many degrees of freedom in the definition of a cast vote. The most natural definition and the one present by default in most state corporate laws (but not in Delaware) would be that it is a proxy sent with either a for or against choice, so that a proposal is approved if votes for represent more than 50% of votes for and against (Threshold 1). More frequent in practice, though, is the case in which proxies sent with an abstain choice are considered as cast and voted against the proposal, in which case a proposal needs to reach 50% of votes for, against, and abstain in order to be approved (Threshold 2), making it harder to pass. More rarely, even broker non-votes and/or shares whose proxy has not 4 In a few cases, the proposal must reach a super-majority threshold (i.e., significantly greater than 50%), always using outstanding voting power as a denominator, but those represent only 1.4% of the proposals in our sample. In the rest of the paper, we rescale voting results for those proposals to make them comparable to simple-majority proposals. 8

9 been sent are considered as voted against the proposal so that a proposal needs far more support in order to be adopted (Threshold 3 and Threshold 4). Even though an approved proposal is not binding, management very often justifies not implementing a proposal by arguing that the proposal did not reach the official bar for approval. Those justifications are often given explicitly after the vote because many third parties push for implementation based on a bar for approval that is uniform across firms- having reached more than 50% of votes cast for and against, and lower than the management approval threshold in a majority of cases. The Investor Responsibility Research Centre (IRRC) has been counting votes and defining majority approval in this fashion ever since it started collecting and publishing voting results in The Council of Institutional Investors (CII), which comprises many of the biggest pension funds, provides every year a list of proposals that reached this level of support to its member institutions. The lists of majority-approved proposals edited by those institutions are then disseminated to even larger audiences through labor unions, online forums and blogs. 6 In our analysis, we focus on three different thresholds. The management threshold is the threshold according to the official voting rule of the corporate charter (which is fixed and known ex-ante, before the voting). The management threshold can essentially be any one of Threshold 1 to 4. The remaining two thresholds of interest are thresholds used by third parties. The Council of Institutional Investors (CII), which represents a large portion of American pension funds, considers a proposal with 50% of votes cast for and against as approved, i.e., it uses Threshold 1. The main proxy advisory firm, Institutional Shareholder Services (ISS), bases its recommendation on 50% of outstanding shares, i.e., on Threshold 4. 7 Figure 2.A. illustrates the different thresholds graphically. Note that the CII threshold is usually (i.e., if there are a positive number of abstentions, non-votes or shares absent to the meeting) lower than the ISS threshold (in terms of required for - votes for approving the proposal) and the management threshold can lie anywhere between these two thresholds or coincide with any of them. 5 IRRC was acquired by ISS in corpgov.net and shareholderforum.com are examples of such online blogs. 7 ISS actually follows a more refined strategy and may trigger some actions when Threshold 1 (CII) is crossed as well. However, as we will explain in more detail in Section 2.3., for our purpose as well as for clarity, we will refer to Threshold 4 when talking about the ISS threshold. 9

10 The bottom line is that management and organizations representing or working for dispersed investors often disagree on whether a shareholder proposal has passed. According to our numbers, this conflict of interpretations arises in about 5% of the cases which, given the large sample of proposals we have in the data, is enough to identify distinct responses of management and shareholder organizations to the voting results. 2.3 What Do Third Parties Do With Voting Results? Disagreements between management and third-party organizations may not bear significant consequences unless some third parties take further actions based on a concept of majority which differs from that of the management. We can identify at least two such responses to the vote in the US context. Ever since shareholder proposals began receiving majority approval, CII has been threatening companies in which a shareholder proposal reaches 50% of votes cast for and against, i.e., crossing Threshold 1, to ask its members to withhold their vote for all management nominees in the next director election if the proposal is not implemented by the board. In order to enforce those sanctions, CII keeps track of the implementation of majority-supported proposals using both public (SEC filings, news releases) and private (requests sent to the board) sources of information. It then issues a list of complying and non-complying firms and makes it available to its members. Those can then individually pick companies in this list and launch vote-no campaigns against them, or the list may serve as an input in the discussions that CII organizes between its members ahead of each proxy season (Anand and Givant Star, 1994). Importantly, CII s sanctions do not make any exceptions: the board as a whole is considered faulty if it has not acted upon the winning vote and CII does not collect individual-specific data on board members so it cannot temper the sanctions for directors who otherwise behaved particularly well. CII itself is a small organization with little staff but many different members (several hundreds) so it is not able to quickly change course when the initial policy reveals itself to be not adapted to a particular case. The main proxy advisor, ISS, may also take sanctions against unresponsive boards. Those consist in issuing recommendations to vote against management nominees at the next director elections. Because many mutual funds blindly follow recommendations made by ISS (Iliev and Lowry, 2014), this may translate into a high number of withheld votes at those elections. There are, however, two key differences with sanctions coming from CII. One is that voting support for a 10

11 proposal needs to reach more than 50% of shares outstanding (i.e., Threshold 4), or it must have reached more than 50% of votes cast for and against (i.e., Threshold 1) at least twice in the last three years. The latter condition represents about 20% of all proposals that reach 50% of votes cast for and against in a given year 8, so in the rest of the paper we focus on the former condition based on outstanding shares. 9 Because this last threshold (Threshold 4) is in most cases distinct from both the CII majority threshold and from the adoption threshold derived from firms bylaws, we can precisely identify the impact of sanctions from ISS. The second divergence from CII guidelines is that ISS s sanctions can be tempered by individual-specific information on nominees so that not all nominees have to be jointly penalized by the lack of response to a well-supported shareholder proposal. This allows us to test whether heavy-handed sanctions on the board from CII have more or less impact than fine-tuned penalties from ISS. 3 Hypothesis Development Given this institutional setting, we now make some detailed hypotheses regarding the impact of votes for shareholder-sponsored proposals. To begin with, voting results might have a discontinuous impact on implementation of proposals: H1a: Passage of a shareholder proposal according to the corporation s bylaws (i.e., voting support being above the Management threshold) increases the likelihood that the proposal is implemented by the board. H1b: Passage of a shareholder proposal according to the Council of Institutional Investors (i.e., voting support being above the CII threshold of 50% of votes for and against (Threshold 1)) increases the likelihood that the proposal is implemented by the board. 8 It could be that markets react on the first strike anticipating that the same proposal will be voted upon the following year and will then trigger the ISS sanctions. However, in unreported results, we do not find any evidence that a proposal is more likely to be included again in the proxy statement or to pass the threshold of 50% of votes for and againstin the following year when it has crossed the threshold of 50% of votes for and against in the current year. 9 Starting in 2013, ISS has recently switched to exactly the same condition as CII: a proposal has to be implemented as soon as it once reaches 50% of votes for and against. ISS s main competitor, Glass Lewis, has issued a similar policy starting in 2013 but with a different threshold: a company must seriously consider a proposal, but not necessarily fully implement it, as soon as it reaches 25% of votes cast for and against. All of those changes took place after our sample period. 11

12 H1c: Passage of a shareholder proposal according to Institutional Shareholder Services (i.e., voting support being above the ISS threshold of 50% of outstanding voting power (Threshold 4)) increases the likelihood that the proposal is implemented by the board. The mechanism for hypotheses H1b and H1c is that the board may want to implement the shareholder-sponsored proposal if the sanctions it faces from CII and ISS at future elections are from its point of view more costly than the implementation of the proposal. Hypothesis H1a relies on some willingness of the board to simply follow the book and wait for an official requirement to consider the proposal before eventually implementing it. Beyond the implementation of the proposal, our description of CII and ISS voting guidelines suggests some direct impact of shareholder votes on shareholder-sponsored proposals on future director elections: H2a: Passage of a shareholder proposal according to the Council of Institutional Investors (i.e., voting support being above the CII threshold of 50% of votes for and against ) increases the likelihood that shareholders withhold their votes for management nominees in future director elections. H2b: Passage of a shareholder proposal according to the Institutional Shareholder Services (i.e., voting support being above the ISS threshold of 50% of outstanding voting power) increases the likelihood that shareholders withhold their votes for management nominees in future director elections. Because CII policies are by design less nuanced than ISS voting guidelines, hypotheses H2a and H2b may have heterogeneous support across management nominees up for election at the same meeting. Our empirical analysis should allow for such heterogeneity. Assuming hypotheses H2a and H2b hold, board members may want to react to the threat of being voted out, so the next natural question is whether one of the board s main prerogatives-monitoring the CEO is affected by CII and ISS potential sanctions following votes on shareholder proposals: H3a: Passage of a shareholder proposal according to the Council of Institutional Investors (i.e., voting support being above the CII threshold of 50% of votes for and against ) increases the likelihood of an abnormal CEO turnover. 12

13 H3b: Passage of a shareholder proposal according to the Institutional Shareholder Services (i.e., voting support being above the ISS threshold of 50% of outstanding voting power) increases the likelihood of an abnormal CEO turnover. By abnormal CEO turnover, we mean turnovers taking place soon after the annual meeting and whose impact on firm value is notably positive or negative. Assuming there is support for any of hypotheses H1, H2, and H3, then comes naturally the question of whether those expected impacts of votes on shareholder-sponsored proposals carry some value for the firm as soon as the voting results are known: H4a: Passage of a shareholder proposal according to the Council of Institutional Investors (i.e., voting support being above the CII threshold of 50% of votes for and against ) increases (or decreases) the stock price following the meeting. H4b: Passage of a shareholder proposal according to Institutional Shareholder Services (i.e., voting support being above the ISS threshold of 50% of outstanding voting power) increases (or decreases) the stock price following the meeting. If there is support for any of those two hypotheses, one should link it with support given to hypotheses H1, H2 or H3. In particular, if there is no support for H1b and H1c, but there is support for either H2 or H3 (CII and/or ISS thresholds matter for director elections, CEO turnover), then support for H4a and/or H4b means that the reduced entrenchment of directors and officers caused by votes on shareholder proposals carries some (positive or negative) value. 4 Data and Descriptive Statistics 4.1 Sample Construction The first segment of our data is on shareholder proposals put to the vote during general assemblies held between 1997 and This dataset is collected by ISS (previously IRRC) and a thorough description of its content is available in Cuñat et al. (2012). One limitation of the dataset is that the voting result variable is consistently reported only according to the CII/IRRC measure of voting support (number of for votes over number of votes for and against ). This is why we add data on voting results from the annual Georgeson corporate governance reviews and 13

14 ISS/Voting Analytics, which measure voting support according to several additional relevant metrics: votes for over votes for, against and abstain, votes for over outstanding voting power. 10 Whenever there is an inconsistency between the three datasets or a missing value for one of the voting metrics, we go back to the SEC filings in EDGAR. 11 Our sample includes the ten most supported proposal types over the period , shareholder support being defined here by the number of times a proposal type has obtained a majority of votes for and against. This leaves us with the following proposal topics (by order of popularity): repeal classified board, eliminate or vote on poison pills, eliminate super-majority requirements, require majority vote for director elections, right to call special meetings, right to act by written consent, vote on golden parachutes, option expensing, say-on-pay, separation between CEO and chairman. The reason we focus on those provisions is two-fold. Firstly, over our sample period, those proposals represent a very large majority (about 90%) of the proposals reaching at least 50% of votes for and against. Other proposal types are far less likely to pass: while they represent about 50% of all governance proposals, only 20% of those will ever go beyond 50% of votes for and against. Because we focus on close-call votes, those numbers mean we do not lose much of either identification power or external validity from excluding the remaining non-csr proposals. Removing proposals that have virtually no chance of passing has also the advantage of improving the statistical power of our tests (Crump et al., 2009). Secondly, focusing on popular proposals greatly helps tracking their implementation: standard proposals are generally more precise and they also elicit quite standard responses from the board. 12 Since we investigate, among other things, the impact of proposals on CEO turnover, the scope of our sampling at firm-level is determined by data availability regarding CEO identity: we must be able to match the proposals dataset with ExecuComp, whose coverage is limited to S&P 1500 companies. This means that our sample comprises 2,404 proposals discussed in 1,878 meetings of 637 different firms. The first row of Table 1 presents the distribution of proposals across time 10 Available on the web: 11 This leads to corrections in about 10% of proposal results in our sample. Results from Georgeson reports are typically the most reliable but they only cover meetings held between January and July, and before 2001 only proposals submitted by non-individuals. 12 Appendix B shows in detail how we proceeded to measure the implementation of proposals. 14

15 (between 1997 and 2011): there are about 160 proposals per year on average. The distribution is hump-shaped with fewer proposals in the early years, peaking in 2003 (after SOX), and fewer proposals in the recent years. Table 2 displays some firm descriptive statistics at shareholder meeting date. The firms we are dealing with are disproportionately big. It is, however, quite natural that the most active shareholder meetings take place in such firms. As shown in Cuñat et al. (2012), these firms also have an important share of institutional investors. Meeting day returns are a good measure of the stock market reaction to the vote because voting results on proposals are typically announced at the end of shareholder meetings. 13 The announcement return at the day of the meeting, while positive on average, is very heterogeneous (e.g., they are -0.90% at the lowest quartile and 1.03% at the highest quartile), suggesting intense information processing by stock markets just after the annual meeting takes place. 4.2 Voting Rules Our data for the vote count rules comes from different sources for different periods. From 1997 to 2006, the ISS/IRRC database documents the exact voting rule, and, from 2003 to 2011, the ISS/Voting Analytics database provides this information. There are, however, multiple inconsistencies either between data sources or within a firm across time, probably often due to ambiguities in the proxy statement itself (Calpers, 2013). Using proxy statements filed in EDGAR before each meeting for those inconsistent cases, we make corrections to about 10% of all proposals in our sample. Table 1 shows also the distribution of voting rules over time. Only in 42.8% of the proposals does the official passing threshold correspond to the majority threshold. In the majority of the cases, abstentions are counted as against votes, i.e., the metric for voting support is the number of votes for over votes for, against and abstain. Default voting rules are defined in corporate state law. 14 However, firms can opt out and change the voting rule through amendments to the corporate bylaws made before the vote takes place. We collect data on the voting rules on state level from LexisNexis. In about 19% of the cases the state 13 Karpoff, Malatesta and Walkling (1996) quote the CEO of the main proxy solicitation firm, Georgeson, as saying that preliminary results are generally given at the meeting, while final tallies are available to investors within hours of the meeting. Since then, the collection of votes is done through the Internet, which most likely makes the announcement of the results even quicker. 14 See Appendix C for a list of voting rules according to each state law. 15

16 rule corresponds to the for over for and against voting metric and in 81% to the for over for, against and abstain voting metric. Companies comply with the state rule in about 78% of the cases; firms are more likely to deviate (25%) if the state rule is based on for plus against plus abstentions compared to only 17% of deviations if the state rule is based only on for plus against votes. For the purpose of our analysis, it is essential that there are many disputed votes; it turns out that the average voting support for a proposal (as a share of votes cast for and against ) is close to 50%, which is comforting. Panel B of Table 1 shows the fraction of proposals that pass the CII, Management and ISS thresholds respectively. On average 46% pass the Management threshold. The passing rate at the CII threshold is slightly higher (51%). Not surprisingly, a lower set of proposals are adopted according to the demanding threshold used by ISS for its sanctions policy. 4.3 Implementation In order to investigate whether those different forms of proposal passage matter for actual implementation of the proposal, we have hand-collected evidence on the implementation of the proposals from various SEC filings accessible on EDGAR. 15 Panel B of Table 1 also shows a dramatic rise in implementation of the proposals conditional on their passage according to corporate charters: the implementation rate went from about 20% before 2002 to more than 70% after Given that at the same time the number of passed proposals has largely increased, this means that shareholder proposals have largely changed in nature after the Enron scandal and Sarbanes-Oxley: they are now clearly an important instrument for changing the way corporate governance rules are set. 16 In Panel B of Table 1, we do not see a clear difference in implementation conditional on passage according to the bylaws and conditional on passage according to ISS and CII, but this is probably because the number of cases where the two diverge is only a small proportion of the whole sample. 15 A detailed list of ways in which firms implement each kind of proposal is available in Appendix B. 16 The trend has already been documented in previous papers (Thomas and Cotter, 2007; Ertimur, Ferri and Stubben, 2010; Renneboog and Szilagyi, 2011) but we are the first to show that the rate of implementation of proposals has reached a very high plateau ever since

17 4.4 Director Elections In order to test our hypotheses related to sanctions taken by CII and ISS following votes on shareholder proposals, we collect data on director elections in the annual meeting following the annual meeting of interest. We obtain the votes on individual directors from ISS/Voting Analytics for the period and 10-Q filings from EDGAR for cases missing in that database and also prior to For each management nominee, we calculate the percentage of votes withheld over votes for and withhold. In the few cases of contested elections, we count as withheld votes that are either effectively withheld or voted for a non-management nominee. We aggregate those votes against directors at firm-level by calculating the minimum and the maximum withholding rate among management nominees for a given election. The minimum withholding rate measures how unpopular (in an absolute sense) is the most popular nominee (relative to other nominees for the same company). We view it as an index of the popularity of the board as a whole rather than just an individual measure of popularity: if even the best nominee from the management receives many defiance votes, it must be that the board as a whole is defied by the electorate. For this purpose, one might think it more appropriate to look at the popularity of the median nominee in terms of election results. The problem is that there are very often just two or four nominees up for election and the median may then be very poorly defined. The mean is not a good choice either as it can be largely influenced by the very large unpopularity of just a few (often one) management nominees and is then not a good measure of how the collective performance of the board is evaluated by the shareholders. One disadvantage of our measure may be that it reflects more the quality of a superstar director than the performance of the board. It turns out however in our data that when there is a nominee whose results stand out from the rest, it is much more often because he is performing very badly in the election rather because he is performing much better 17. For that same reason, the maximum withholding rate is quite likely to reflect a truly individual-specific assessment of the nominee who receives this relatively low level of support rather than an assesment of the board as a whole, and we will use it as such in the rest of the analysis. 17 One can readily see this from table 2, panel D, as the mean withholding rate is significantly closer to the minimum withholding rate than it is to the maximum withholding rate. 17

18 Withholding rates are relatively small (around 9% of the votes withheld on average according to Table 2), but it is generally recognized that even small yet significant levels of defiance at director elections can be considered a blow to incumbent board members and management (Cai, Garner and Walkling, 2009). For example, when 23% of votes for AOL s chairman Steve Case were withheld, he felt compelled to resign. Therefore, a large percentage increase in withheld votes following the passage of a proposal may be sufficient to trigger strong decisions from the board. It is important to note as well that the distribution of withholding rates is not normally distributed as there is a mass of firms with very low withholding rates but also a significant number of firms with large withholding rates. This is why we will also consider the logarithm of the withholding rate for nominees in the rest of our analysis 18. Voting Analytics also provides us with the voting recommendations given by ISS for about 85% of the director elections in our sample. Prior to 2003, the database only mentions whether all nominees are recommended to be voted for, all votes for nominees are recommended to be withheld, or some should be withheld and some should not. We aggregate the recommendations at the firm-meeting level and construct a dummy indicating whether there is at least one recommendation to withhold and a dummy indicating whether ISS issues a Withhold recommendation for all management nominees. We cannot precisely link a recommendation to the amount of defiance votes he receives. However, using the whole sample of director elections from ISS/Voting Analytics, we find that in 95% of the cases when ISS issues just one withhold recommendation it is actually going to the nominee who will eventually receive the lowest level of support in the election. This means that one can estimate the effect of a bad recommendation on voting results by just looking at whether or not the existence of at least one withhold recommendation translates into a significantly lower level of support for the most unpopular candidate. Recommendations to withhold are more frequent than what the actual withholding rates in director elections would suggest but they are still a small minority of the recommendations (about 27% 18 The non-normality of withholding rates is less true when the ex-ante degree of shareholder defiance towards management nominees is high. Therefore, it is to be expected that absolute withholding percentages can be more precisely estimated (and thus taking the logarithm less necessary) as the past level of voting support for a proposal increases. This will matter in our discussion of the results for director elections. 18

19 according to Table 2), which fits the available evidence on proxy advisor recommendations (Cai, Garner and Walkling, 2009). 4.5 CEO Turnovers We want to test the effect of the vote on shareholder proposals on the quantity and quality of CEO turnovers. To this effect, we identify turnovers using changes in the identity of a firm s CEO in Execucomp. Using this information, we obtain several pieces of information on each turnover from newspaper articles and press releases collected in Factiva: announcement date, whether the turnover is an integral part of a merger, whether the turnover is caused by death or severe health problems of the CEO. We focus our attention on the first turnover announced less than two years after the day of the shareholder meeting. We choose this period of time because CEOs may leave in anticipation of the next director elections (i.e., between t and t+1) or due to the results obtained by incumbents in those elections (i.e., between t+1 and t+2). If a turnover has been announced before the meeting but has not yet taken place effectively by that date, we only consider as a turnover event the one that was announced for the CEO-elect at the time of the meeting. We also restrict our attention to departures unrelated to a merger or bad health. Given the small likelihood of such events (around 10% of all turnovers), none of our results are affected by their inclusion in our sample. After all these data filters, we find that 530 CEOs have announced their departure less than two years after a shareholder meeting in which one of our sample proposals has been put to the vote. This means that shareholder meetings are followed by a CEO turnover in the next two years in 22% of the cases. For each turnover, we compute announcement abnormal returns using the market model. Descriptive statistics for the sample of CEO turnovers are in Panel C of Table 2. CARs on turnover announcement are centered around zero. This does not mean however that these are irrelevant events: the standard deviation of abnormal returns more than doubles on the day of announcement. This simply suggests that turnovers have very heterogeneous value implications across firms: some are viewed as negative events and some as positive. For that reason, just looking at the effect of governance on the frequency of turnovers might entail a substantial information loss and one should instead consider CEO departures that affect firm value positively (the good turnovers) and those that impact stock prices negatively (the bad turnovers) as distinct outcomes. Because only a quarter of turnover announcement CARs are below minus 19

20 1.87%, we define all such unambiguously value-decreasing turnover events as our main outcome of interest in the rest of the paper. One characteristic that distinguishes such bad events from other turnovers is that the leaving CEO is then younger than usual. This would fit the idea that bad turnovers are such in part because they happen too early in the relationship between the firm and the CEO. 5 Empirical Strategy As discussed above, shareholder organizations, proxy advisors and management are all reacting to the voting support received by a shareholder proposal in a discontinuous way: votes for must go above a certain threshold before each of those players sharply changes its behavior. To the extent that voting support for a proposal has a significant random component, this allows us to use a Regression Discontinuity Design (RDD) in order to tease out the causal effects of each player s reactions to the vote. 5.1 RDD Estimation The RDD methodology brings a series of concerns about estimation and its efficiency. The reason is that, unless the sample of close-call votes has infinite size, there are never enough instances in which vote shares are just at one and the other border of the passing threshold to guarantee a reasonable level of statistical power for tests of the significance of the impact of proposal passage. With a finite sample, it is therefore necessary to use information far away from the threshold and compensate for the potential extrapolation bias by modelling the continuous relationship between the vote share and the outcome on each side of the treatment threshold. This modelling process contains a lot of degrees of freedom, but the econometric literature has converged towards a set of best practices which we will closely follow (Lee and Lemieux (2010)). For all of our outcomes and treatments of interest, we begin with plotting averages of the outcome of interest over small intervals of the voting metric according to which the passing threshold is defined. Those bin averages are essential because they give a sense of the credibility of the jump caused by crossing the majority threshold. In order to be convincing those intervals over which averages are constructed should be small enough that the link between the outcome and the vote looks quite erratic, but big enough to make sure that a jump at the treatment threshold is visible if 20

21 it is really there. We follow the procedure laid out in Calonico, Cattaneo and Titiunik (2014) in order to produce graphs which respect those two conditions. Those graphs come together with a global polynomial fit of the data to the right and to the left of the passing threshold in order to get a sense of a magnitude of the treatment effect. Once reassured by non-parametric graphs about the existence or lack of an effect of proposal passage, we turn to local linear regressions in order to provide precise estimates and statistical tests. The key parameter in such an exercise is the bandwidth of the non-parametric estimation; it should be small enough to limit the extrapolation bias and big enough to provide statistical power. We use the algorithm designed by Imbens and Kalyanaraman (2012) to obtain such an optimal bandwidth. We provide treatment estimates for bandwidths that are 50% lower and 50% bigger than the optimal level so as to make sure results are not overly sensitive to bandwidth choice. Another robustness check consists in using instead a parametric approach: we compare means of the outcome three percentage points to the left of the majority threshold and three percentage points to the right. This has the advantage of being intuitively the closest equivalent to considering the passage of a proposal as a random event (Calonico, Frandsen, Titiunik, 2014). 5.2 RDD with Multiple Majority Thresholds One specific challenge of our design is that there are multiple treatment thresholds as voting support for a proposal grows. When those thresholds are purely identical, i.e., when corporations bylaws use the same majority concept as either CII or ISS, it is impossible to disentangle effects coming from management actions and effects coming from third parties responses to the vote. When each party uses different voting metrics, the RDD method does not in principle need any adjustment: when an observation is at the border of one majority threshold, it cannot simultaneously be at the border of another kind of majority threshold. In practice, sample size is limited and, as discussed above, some extrapolation using observations far from the treatment threshold is required. For that reason, the treatment effect of proposal adoption according to one majority requirement may be contaminated by the treatment effect of proposal adoption according to another majority requirement. In order to deal with this problem, we test our hypotheses using different specifications and subsamples: 21

22 Specification 1: Mngt-CII > 0.5% o In our main specifications, we focus on proposals where the management threshold is different from the CII threshold. Moreover, we require that the number of votes counting against the proposal according to the CII and management rule differ by at least 0.5% of votes effectively cast for and against. 19 We thus eliminate proposals for which the voting metrics strictly overlap (42.8% of all proposals) and those for which the gap between voting metrics is less than 0.5% (8% of all proposals). We choose this threshold of 0.5% because in unreported results we find that the probability of implementation jumps discontinuously according to RDD estimations in the sample of proposals with a gap between voting metrics smaller than 0.5% but not at all when the gap is bigger than 0.5%. Choosing a larger gap significantly reduces the number of observations and statistical power. We estimate the treatment effect non-parametrically and report estimates with the optimal bandwidth as well as with bandwidths that are 50% larger or smaller, respectively. Figure 2.B.1 illustrates this sample restriction. Specification 2: Mngt-CII > 2% o Our second specification differs from specification 1 by requiring the vote share difference between management and CII rule to be at least 2%. We use this specification for two reasons. One is to check for the sensitivity of our baseline results to changing the sample restriction used in our baseline regressions. Another reason is specific to evaluating the stock market reaction to the vote. When the gap between voting metrics used by CII and management is very small, it is hard to believe that markets are able to distinguish approval according to CII and approval according to management very quickly. For that reason, the instant market reaction to voting results when there is a conflict of interpretations between CII and management is likely to be muted if the actual gap between their voting metrics is 19 As an example, consider the case when the management rule is to count abstentions in the denominator (which is the same thing as treating them as votes against the proposal) and the proposal obtains the following results: 501 for, 499 against and 50 abstentions. The voting result is 50.1% according to CII, 47.7% according to management, and the corresponding voting gap is 5% (i.e.. 50/( )) of votes for and against. 22

23 very small. 20 Since Cunat et al. (2012) show that markets are able to predict voting results just before the vote takes place with a margin of error of little more than 2%, it is a fortiori reasonable to expect that markets are able to quickly distinguish approval concepts just after the vote took place when the gap between voting metrics is larger than 2%. Given this tighter restriction, we expect to reduce the statistical power of our tests, as we then keep only about a quarter of all proposals. Figure 2.B.2 illustrates this sample restriction. Specification 3: Full Sample o Our third specification employs the full sample, i.e., it also includes proposals where management rule and CII rule coincide. The treatment effect is estimated nonparametrically, using the optimal bandwidth. Figure 2.B.3 illustrates this sample restriction. This specification has the advantage of having the greatest external validity. However, one should remain cautious about the interpretation because those estimates often measure the combined reactions of both CII/ISS and management. Specification 4: No Contamination Sample o We also use as robustness checks samples in which the treatment of interest varies but the other treatment statuses remain constant. The most important of such no contamination samples is the one where we only consider proposals that are never passed according to management in order to estimate the causal effect of a proposal obtaining a majority according to CII. Figure 2.B.4 illustrates this sample restriction. This restriction leaves us with 31.4% of all proposals. Specification 5: Parametric estimation with controls 20 For that reason, specification 2 will be very important when investigating the effects of passing the thresholds on valuation (see Section 6.5)). 23

24 o We also estimate the treatment effect parametrically using OLS around the majority threshold (+/- 3%). 21 Our main independent variable is a dummy variable that is equal to one if the proposal has passed the approval threshold of interest. We control for potential contaminating effects of crossing / not crossing other thresholds by including passing dummies of all thresholds in the regressions. Figure 2.B.5 illustrates this sample restriction. Using a Regression Discontinuity Design requires sufficiently many observations close to each of those thresholds. Table 3 also shows the number of proposals with voting outcomes close to each of the three thresholds. (47, 53) corresponds to the number of proposals with voting outcomes between 47%-53% using the respective counting rule. We have about 322 and 337 proposals in our sample when applying a -3/+3% range around the the CII and Management Threshold, respectively; numbers are lower around the ISS threshold as it is more unlikely that proposals reach these levels of support (the corresponding number is 224). 5.3 Internal Validity of the Voting Experiment The internal validity of our estimates rests on the assumption that small variations in the vote share obtained by a proposal are essentially random. This means in particular there is not any systematic manipulation of the results when the result is a close call. This kind of assumption makes perfect sense in the context of general elections but it has been shown to fail in contexts in which voters are few and may vote strategically such as roll-call votes in US Congress (McCrary, 2008). Such a possibility can in fact be properly tested in our sample. One such test has been proposed by McCrary (2008) and rests on the assumption that if there was strategic voting, one should observe that the density of proposals subject to a vote exhibits a significant jump at the 50% majority threshold. In Figure 3, we provide a non-parametric graphical version of such a test, using our baseline sample restriction. For each one-point interval of proposal vote shares, we count the number of proposals put to the vote in our sample. The corresponding density test, as suggested by McCrary, estimates two local linear regressions, to the left and right of the cutoff point. One other test of the randomness of passing a governance proposal in closely-contested votes consists in 21 We follow the parametric methodology suggested by Calonico, Frandsen, Titiunik (2014) and we choose the -3/+3 window because it is the largest window such that difference-in-means tests below and above the thresholds are insignificant for pre-treatment outcomes (see Table 3). 24

25 running placebo experiments with outcomes that cannot possibly be affected by the passing of the proposal because they were measured before the meeting. In Table 3, we run our RDD local linear regressions as well as a difference-in-means estimation on a series of such past outcomes. We start by analyzing the CII threshold. Figure 3.A shows that there is no significant jump in density at the majority threshold, so our claim that voting is random around that 50% zone passes a first important test. The estimated log discontinuity from the corresponding McCrary test is and not statistically significant from zero. Table 3 analyzes the effect of passing the CII threshold on some previously determined outcomes: market capitalization, Tobin s Q, CEO wage, CEO age, and corporate governance (G-index). None of the estimated effects are statistically significant. Running similar tests on the ISS threshold (Table 3, panel B, and Figure 3.B), we do not find any evidence of manipulation either (the estimated log discontinuity is 0.06 and not statistically different from zero). 5.4 External Validity of the Voting Experiment One last caveat of our identification approach is common to all regression discontinuity designs and it is that our causal estimates may not be representative of a significant share of US companies. To this effect, in Table 3, panel A and B, we compare a series of structural firm-level variables measured prior to the meeting in the total sample and in the sample of closely-contested proposals (between 47% and 53% of the vote share). Analyzing the CII threshold, we find that closely-contested proposals take place in slightly bigger firms, with a difference of about 20% on average, but those firms have no specific pattern in terms of performance (as measured by Tobin s q). Incumbent CEOs of firms with closely-contested meetings do not exhibit distinct personal characteristics, which comfort us in the belief that our estimates of the effect of shareholder organizations on director elections and CEO turnovers will carry a high level of generalizability to other contexts. The picture looks quite different at the ISS threshold. CEO characteristics such as age and firm profitability are not significantly different in firms whose proposals reach this high a level of support. However, those firms tend to be much smaller and also slightly less well governed. Not surprisingly then, CEOs of firms with close contests according to ISS have a much higher wage ratio, measured pay over market cap. Those facts are easy to rationalize: it is harder for a proposal 25

26 to reach 50% of outstanding shares if the number of shareholders that must be voting is very large, unless the firm is particularly badly-governed. In principle, the degree of entrenchment of management is higher in such firms, so it should also be where the impact of ISS recommendations on turnovers, director elections and value should be the highest. 6 Results 6.1 Implementation It is natural to start the analysis by looking at the impact of the vote on the implementation of the proposal being voted upon. Indeed, all three key majority thresholds we have identified may sharply increase the likelihood of implementation: crossing the CII and ISS thresholds may trigger punishment of the board of directors by those institutions if the board does not implement the proposal; if the Management threshold is crossed the bylaws of the firm force the board to consider the implementation of the proposal. Figures 4 and Table 4 present our results for implementation. In our figures and baseline regressions, we exclude cases where voting rules used by CII or ISS and by the Management are identical or almost identical as we are interested in analyzing differential effects of crossing each of these thresholds. The graphs provide stark visual evidence that it is only the Management threshold that matters for implementation: Figure 4 show that crossing the CII threshold does not lead to a discontinuity in the likelihood of adopting a proposal. While the likelihood of adoption is very flat and very low (around 10%) for low shareholder support, it raises continuously with higher support of shareholders. The picture looks similar when looking at the threshold considered by ISS. Given that the ISS threshold is based on shares outstanding, being close to that threshold implies high shareholder support in terms of any other threshold and hence, we do not see the flat part of the corresponding figure for the CII threshold. The picture looks very different when analyzing the management threshold in Figure 4. The implementation likelihood is also relatively flat for low levels of shareholder support and is rising as well. However, when the 50%-threshold is passed, there is a very sizeable jump of more than 20% of the implementation likelihood (around 20% below the threshold compared to more than 40% above the threshold). Corresponding regressions (Table 4, Panel B) show that the likelihood of implementing the proposal increases between 16%-32% once it has been approved according to the bylaws. 26

27 The effect of a proposal being approved according to the CII and ISS definitions in most specifications is negligible, both economically and statistically. One exception in Table 4, Panel A is the specification that employs the full sample (column 5): the estimate for the effect of crossing the CII threshold on implementation is large (15.72%) and significant at the 1% level. However, the problem is that in many cases, CII and management threshold lie close to each other or are even identical (see discussion in Section 5.2). The effect disappears when we look at more carefully designed specifications (see column 1 to 4 and 6 to 7) that are motivated to address this problem. In columns 1 to 4 we only look at proposals in which the management threshold is not the CII threshold and there is a certain distance between them (0.5% and 2% of the votes), column 6 uses the no contamination sample, and column 7 uses only proposals that are close to the threshold. Overall we provide evidence that boards implement shareholder proposals when the bylaws nudge them to do so, not when they face sanctions from proxy advisors and shareholder organizations. This may be because ISS and CII s actions have no impact on them, something we are going to test in the next sub-sections, or because management is privately valuing the proposal content very negatively and would rather risk being sanctioned by CII and ISS than actually implement the proposal. 6.2 Director Elections Does this mean that the assessments of the voting results made by CII and ISS simply do not matter? To answer this question, we estimate the impact of shareholder proposal votes on next year s director elections. We posit that sanctions from ISS and CII might affect nominees differentially depending on their popularity, which is why we look at impacts on defiance votes received by the nominee with the most and the least defiance votes received at the meeting. Given that CII does not sanction individual directors but only the whole board and that the lowest rate of withheld votes among management nominees of a company is more likely to reflect collective board performance (as discussed above), we expect to see that CII sanctions affect the rate of support received by the most popular nominee in each company rather than the level of support received by the least popular candidate (whose popularity may be so low that the broad-based CII sanctions do not increase the number of shareholders voting against him). On the contrary, ISS makes nominee-specific recommendations and for that reason rarely sanctions an entire set of candidates. Because the ranking of candidates by ISS is very correlated with the relative levels of 27

28 support they eventually receive in the election (Cai, Garner & Walkling, 2009), we expect that the most popular candidate is the least likely to be sanctioned by ISS after the board refuses to implement a majority-supported proposal as defined by ISS. Figures 5 to 7 and Table 5 present the results of our tests. As soon as a proposal is approved according to CII, there is a significant jump in the number of votes against management nominees the following year. Interestingly, this does not affect all nominees equally. The nominee who fares best in the election in relative terms receives a much more significant amount of votes against him in absolute terms (see Figures 5.A and rows 1 and 2 of Table 5, Panel A). The best director receives about 75% (the corresponding coefficient is 0.56) more votes against her. The latter effect translates into a large number in terms of vote shares (minus 5.74 points). This is even more impressive as existing literature has convincingly shown that even small movements in withholding rates might significantly affect the board s decisions (Cai, Garner and Walkling, 2009). Note that even though the results in row two are not statistically significant, the estimated treatment effects are large in magnitude and negative across all specifications 22. In contrast, the nominee who fares the worst in the election is in fact barely affected (see Figures 6.A and rows 3 and 4 of Table 5, Panel A), suggesting shareholders just do not wait for a CII recommendation before voting him out. That the best director suffers more from the approval of a proposal according to CII than the worst director shows that CII sanctions affect the board as a whole with little attention paid to individual situations. This was to be expected since CII does not collect individual-specific information on the board actions. Unsurprisingly, ISS director ratings are affected by voting support for a shareholder proposal the previous year if and only if the proposal has reached more than 50% of outstanding voting power (see Figures 7 and rows 1 to 2 in Table 5, Panel B). This impact is asymmetric: the chance that there is at least one withhold recommendation increases by percentage points but the chance that ISS only issues withhold recommendations increases by only 8.68 percentage points because 22 That the absolute withholding rates estimates are not statistically significant was to be expected as withholding rates are quite close to 0% around the CII threshold so the distribution of absolute withholding rates is clearly non-normal in that range (as opposed to the logarithm of withheld votes). 28

29 of the proposal passage as defined by ISS guidelines. This suggests that ISS often uses nomineespecific information to protect the best nominees against its all-or-nothing sanctions policy. An important question is whether those additional Withhold recommendations translate into higher withholding rates in director elections. Figures 5.B and 6.B show that it is only the nominee receiving the least votes in her favor whose amount of withhold votes significantly increases due to shareholder proposal reaching the ISS support threshold. Regressions (rows 3 to 6 in Table 5, Panel B) indicate that crossing the ISS threshold increases withheld votes against the least popular nominee by about 26%, which corresponds to a decline in share of voting support of more than 4 points. Using the ISS threshold as an instrument, one can perform an IV regression 23 of the impact of receiving at least one Withhold recommendation on the voting support received by the least popular nominee. The effect is strong: while the existing literature estimates an effect located between 10% and 30% with strong suspicions of an upwards endogeneity bias (Cai, Garner and Walkling, 2009; Choi, Fisch and Kahan, 2010; Li, 2013), we estimate that a withhold recommendation causes the share of voting support for the least popular nominee to decline by 26 points on average across specifications (rows 7 and 8 of table 5, panel B), which corresponds to a multiplication of the number of votes against by about five. 24 The most popular management nominee is barely affected by a proposal reaching a level of support big enough to trigger ISS sanctions. This is in great part due to the fact that, as evidenced above, ISS tones down its sanctions against nominees with a particularly good track record, but may be also due to the fact that ISS clients may not follow recommendations as much when they appear too heavy-handed. 6.3 CEO Turnovers One way of documenting the pressure then put on boards is to look at CEO turnover. We start our analysis by testing whether the voting outcome on shareholder-sponsored proposals leads to a higher probability that a good CEO leaves a firm. We define the quality of the leaving CEO in terms of how much value is lost upon the announcement of his departure. At the CII threshold, there is a large positive jump on the frequency of bad turnovers (Figure 8), i.e., those that lead to a negative stock market reaction upon turnover announcement. The likelihood of a bad turnover 23 To be more specific, we use a fuzzy regression discontinuity design, with an optimal bandwidth specifically designed for this purpose (see Imbens and Kalyanaraman (2012)). 24 It is well known that fuzzy designs lead to greater dependence on bandwidth choice, as one can judge by our results. For that reason, it is reasonable not to pick one preferred estimate but an average across specifications. 29

30 goes from around 8% to about 30% once voting support for the proposal crosses 50% of votes for and against (see Figure 8 and Table 6, Panel A, row 1). We do not find any consistently significant effect of majority approval according to CII for medium or good turnovers (Table 6, row 2 and 3). Interpreting stock market reactions to CEO turnovers suffers from ambiguities because a board which does its job by collecting private information on the CEO and taking its decision to fire the CEO based on that piece of information will inevitably reveal some bad news about the firm s prospects at the same time that it is taking a value-enhancing decision for the firm. However, if this were the case, this should often show up in the accounting performance results prior to the firing decision (i.e., the fired CEO consistently underperformed). This provides a simple test in which we distinguish turnovers that are not well received by the market based on the performance of the firm in the years preceding the turnover as well as in the years following the turnover. Following Denis and Denis (1995), we define pre-turnover operating performance as the evolution of the ROA from t-3 to t-1. In order to control for mean-reversion, industry-and-year effects, we follow the approach by Barber and Lyon (1996): we match each firm in t-4 to a group of firms in the same SIC2 industry and in the same decile of ROA, then for all subsequent years, we subtract the median ROA in this control group from the raw ROA. Once we have defined this industry-and-performance-adjusted ROA, we group turnovers depending on whether this ROA measure increased or decreased before the turnover. We can then use our RDD methodology to look at the causal effect of governance on each kind of turnover. The results are displayed in rows 4 and 5 of Table 6, Panel A. At the CII threshold, we do not find that proposal passage specifically increases the probability of turnovers with a negative stock market reaction and following a bad accounting performance; in fact it is slightly more likely (albeit not statistically significant) that there is a CEO turnover following a good operating performance of the firm and triggering a negative stock market reaction upon announcement. Assuming that boards private information on performance is positively correlated with observable accounting performance, this is at odds with the story that boards use their information to make sound firing decisions and more in line with a story in which it is good CEOs rather than bad ones that leave firms when board members feel pressured by the majority passage of a proposal according to CII. 30

31 The above results are in line with the conclusion from our analysis of director elections that sanctions taken by CII can lead to inefficiencies ex post. In order to check whether this is due to the fact that those sanctions are too coarse, we now turn to the analysis of the impact of ISS voting guidelines (Table 6, Panel B and Figure 8.B). The results are very clear: there is simply no effect of having a shareholder proposal go beyond the adoption threshold set by ISS on CEO turnovers, be they value-enhancing or value-destroying. This needs to be put in relation with the previous finding that ISS majority vote policies primarily affect the popularity of the least popular directors, as good directors are saved by their otherwise good track record. Those safe talented directors may then be able to make sure that no inefficient CEO turnover decision is taken. This counter-example confirms that it is most likely the coarseness of CII penalties in case of non-compliance by boards which is the source of value-destroying CEO turnovers. 6.4 Valuation Our previous tests show that adoption of proposal according to the bylaws triggers implementation, while adoption according to CII (resp. ISS) strongly (resp. weakly) reduces the entrenchment of boards and CEOs. It is not clear whether any of those consequences are positive or negative for firm value, which is why we now turn to the analysis of the stock reaction to voting results on meeting day (Figures 9, and Table 7). We find a significant increase in the stock price as soon as a shareholder proposal reaches 50% of the votes for and against (i.e., the adoption threshold set by CII), with an effect of around +0.79% (column 1 in Table 7, Panel A). The estimated effect slightly increases (1.19% %) in the specifications that aim to cope more efficiently with a potential contamination of the management threshold. As discussed in detail in Section 5.2 shareholders need to be able to quickly understand whether certain thresholds have or have not been crossed in order for us to detect a stock market price reaction. Therefore, we believe it is very important to focus particularly on column 4 (i.e., Specification 2) where we require a sufficiently large gap between the ISS threshold and the management threshold. Our interpretation of this abnormal return is, however, quite distinct since the only thing that changes at this threshold is not the likelihood of implementing the proposal but instead the stark reduction in CEO and board entrenchment due to CII sanctions of unresponsive boards. The 31

32 positive abnormal return shows in fact that markets value this dis-entrenchment in spite of some ex post inefficiencies revealed by the following increase in value-decreasing CEO turnovers. The fact that, as opposed to CII sanctions, markets simply do not react when suddenly sanctions from the proxy advisor ISS are more likely (Table 7, Panel B) suggests further that those ex post inefficiencies linked to undue pressures exerted by CII on talented board members and CEOs are actually a good disciplining tool ex ante. 6.5 Do The Effects of the Vote Depend on Actual Implementation? The evidence presented so far shows that the largest part of the impact of shareholder support for a shareholder-sponsored proposal comes from crossing thresholds of adoption defined by the third parties CII and ISS. We have proposed that those causal effects arise because then CII and ISS issue penalties to the board if it does not implement the proposal. Because in our sample there is a significant rate of implementation of proposals, our previous estimates may underestimate the true causal impact of ISS and CII sanctions. The implementation rates in our sample for proposals close to the CII and ISS thresholds are 31% and 61%, respectively. This means the impacts of approval according to CII and ISS should be multiplied by 1.45 (=1/(1-0.31)) and 2.56, respectively, if one wants to get at the causal effect of receiving a sanction from those two organizations: 25 being targeted by CII increases withheld votes of good directors by 126%, bad CEO turnover by about 30%, and firm value by 1.15%; ISS sanctions increase withheld votes of bad directors by 80%. 26 However, this kind of computation is only valid if reaching the thresholds of 50% of votes for and against and 50% of outstanding voting power matters only due to conditional sanctions delivered by ISS and CII and not instead due to some other mechanism affecting the firm regardless of the actual (non-)implementation of the proposal. Proposals reaching those levels of support might indeed cause real effects even before we know whether they will be implemented if this increases the media exposure of the firm in the days, weeks and months between the annual meeting and the time at which the decision to implement the proposal or not is taken. 25 Our measure of implementation most likely suffers from some measurement error because we do not observe the assessments made by ISS and CII of the efforts made by the firm to implement the proposals. This means our results still understate the size of the impact of CII and ISS sanctions. 26 For instance, the effect of 126% on withheld votes for good directors is calculated as follows: We first scale our baseline estimate of 0.56 by the mentioned multiplier 1.45 to obtain a scaled coefficient of We then calculate the semi-elasticity computing exp(0.82)-1. 32

33 To answer this concern, we present in Table 8 our main estimations run so far, only now in two subsamples: one in which eventually the proposal is implemented in the following year and one in which it is not. Conditioning the analysis of the impact of the vote on a variable determined after the vote is not problematic here because we look at the impact of for votes crossing the CII and ISS thresholds of approval and we have convincingly shown above that those events are in fact orthogonal to the implementation decision. Rows 2 to 5 of Panels A and B of Table 8 analyze outcomes after the meeting that have previously been found to be affected by passing the CII and ISS threshold, respectively. The results show that virtually nothing happens around those thresholds of voting support for the proposal when it is eventually implemented, while the estimates are even larger than in the pooled analyses (Tables 4 to 7) and highly significant when the proposal is eventually not implemented. Most interestingly, the market even seems to be able to anticipate this non-implementation and the value of the ensuing punishment exerted by CII: The first row of Panels A of Table 8 shows that a majority votes trigger a far smaller reaction in the stock price on meeting day when the proposal is eventually implemented (+0.39%) rather than not (+1.03%). 27 The difference becomes even more striking when we restrict the sample to proposals where the stock market is more likely to distinguish the CII and Management threshold more clearly. 28 The corresponding effects are estimated to be 2.01% when not adopted versus 0.30% when adopted. A similar pattern arises with the ISS threshold: all our average effects of crossing this threshold disproportionately come from cases where the firm eventually did not implement the proposal. In other words, the impact of voting results on the various outcomes we have analyzed can indeed be fully ascribed to the CII and ISS policies consisting in pointing the finger towards unresponsive boards. 7 Conclusion In this paper, we provide a new mechanism for how shareholder voting on non-binding shareholder proposals can have real effects on the firm: a large part of the impact does not come from the actual 27 The true impact of the sanctions on the stock price is underestimated because stock markets cannot perfectly anticipate the decision to implement conditional on the voting result. 28 See our discussion in Sections 5.2. and

34 implementation of those proposals. Rather, votes very often matter only because they are used as an input by activist shareholder organizations and proxy advisors to put additional pressure on otherwise entrenched boards. This result matters because it means that board members and CEOs cannot fully control the consequences of shareholder voting: even when its de jure consequences are negligible, the vote might still give enough information to activist organizations that something wrong is going on with the firm that deserves exerting significant pressure on corporate insiders. We also show that this kind of low-cost activism is valuable not in spite of but because of the coarseness of its mode of action. This implicitly means that many boards of US firms are currently very entrenched, to the detriment of investors. Current efforts to impose larger proxy access, which would automatically reduce the level of entrenchment of the board at a small cost, may be the ultimate solution to this problem. Analyzing the long-run impact of this reform is an important topic for future research. 34

35 8 References Allen, C. H. (2007). Study of Majority Voting in Director Elections, Technical report, Neal, Gerber & Eisenberger LLP. Anand, V. and Givant Star, M. (1994). Protest considered by CII - Members may withhold support from some directors, Pensions and Investments, 21 February Barber, B. and Lyon, J. (1996). Detecting abnormal operating performance: The empirical power and specification of test statistics, Journal of Financial Economics 41, pp Bebchuk, L. (2007). The Myth of the Shareholder Franchise, Virginia Law Review 93, pp Bebchuk, L., Cohen, A. and Ferrell, A. (2009). What Matters in Corporate Governance?, The Review of Financial Studies 22, pp Becker, B. and Subramanian, G. (2013). Improving Director Elections, Harvard Business Law Review 3, pp Black, B. (1998). Shareholder activism and corporate governance in the United States, New Palgrave Dictionary of Economics and the Law 3, pp Cai, J., Garner J., and Walkling R. (2009). Electing Directors, Journal of Finance 64, pp Cai, J., Garner J., and Walkling R. (2013). A paper tiger? An empirical analysis of majority voting, Journal of Corporate Finance 21, pp Calonico, S., Cattaneo, M. and Titiunik, R. (2014). Optimal Data-Driven Regression- Discontinuity Plots. Working paper, University of Michigan. Calpers (2013). Vote Calculation Methodologies, technical report. Cattaneo, M., Frandsen, B. and Titiunik, R. (2014). Randomization Inference in the Regression Discontinuity Design: An Application to Party Advantages in the U.S. Senate, Journal of Causal Inference, forthcoming. 35

36 Cheng, I., Hong, H. and Shue, K. (2014). Do Managers Do Good with Other People's Money?, Working paper, University of Chicago Choi, S., Fisch, J. and Kahan, M. (2010). The Power of Proxy Advisors: Myth or Reality?, Emory Law Journal 59, pp Crump, R., Hotz, J., Imbens, G. and Mitnik, O. (2009). Dealing with limited overlap in estimation of average treatment effects, Biometrika 96, pp Cuñat, V., Giné, M. and Guadalupe, M. (2012). The Vote Is Cast: The Effect of Corporate Governance on Shareholder Value, Journal of Finance 67, pp Cuñat, V., Giné, M. and Guadalupe, M. (2013). Corporate Governance and Value: Evidence from Close Calls On Shareholder Governance Proposals, Journal of Applied Corporate Finance 25, pp Cuñat, V., Giné, M. and Guadalupe, M. (2014). Say Pays! Shareholder Voice and Firm Performance, LSE working paper. Cvijanović, D., Dasgupta, A. and Zachariadis, K. (2014). Ties that Bind: How Business Connections Affect Mutual Fund Activism, ECGI working paper. Del Guercio, D., Seery, L., and Woidtke, T. (2008). Do boards pay attention when institutional investor activists just vote no?, Journal of Financial Economics 90, pp Denis, D. & Denis, D. (1995). Performance Changes Following Top Management Dismissals, Journal of Finance 50, pp Ertimur, Y., Ferri, F. and Oesch, D. (2014). Does the director election system matter? Evidence from majority voting, Review of Accounting Studies, forthcoming. Ertimur, Y., Ferri, F. and Stubben, S. (2010). Board Of Directors Responsiveness To Shareholders: Evidence From Shareholder Proposals, Journal of Corporate Finance 16, pp Ferri, F. (2012). Low-Cost Shareholder Activism: A Review of the Evidence, in Research Handbook on the Economics of Corporate Law, ed. C. A. Hill and B. H. McDonnell., pp

37 Fos, V. (2013). The Disciplinary Effects of Proxy Contests, UIUC working paper. Gillan, S. and Starks, L. (2000). Corporate governance proposals and shareholder activism: the role of institutional investors, Journal of Financial Economics 57, pp Gillan, S. and Starks, L. (2007). The Evolution of Shareholder Activism in the United States, Journal of Applied Corporate Finance 19, pp Giné, M. and Moussawi, R. (2007). Governance Mechanisms and Effective Activism: Evidence from Shareholder Proposals on Poison Pills, Working paper, University of Pennsylvania Grundfest, J. (1993). Just Vote No: A Minimalist Strategy for Dealing with Barbarians inside the Gates, Stanford Law Review 45, pp Illiev, P. and Lowry, M. (2014). Are Mutual Funds Active Voters?, Review of Financial Studies, forthcoming. Imbens, G. and Kalyanaraman, K. (2012). Optimal Bandwidth Choice for the Regression Discontinuity Estimator, Review of Economic Studies 79, pp Kahan, M. and Rock E. (2008). The hanging chads of corporate voting, Georgetown Law Journal 96(4), pp Karpoff, J. (2001). The impact of shareholder activism on target companies: a survey of empirical findings. Working paper, University of Washington, WA. Karpoff, J., Malatesta, P.. and Walkling R. (1996). Corporate governance and shareholder initiatives: Empirical evidence, Journal of Financial Economics 42, pp Lee, D. and Lemieux, T. (2010). Regression Discontinuity Designs in Economics, Journal of Economic Literature 48, pp Levit, D. and Malenko, N. (2011). Nonbinding voting for shareholder proposals, Journal of Finance, 66(5), Li, T. (2013). Outsourcing Corporate Governance: Conflicts of Interest and Competition in the Proxy Advisory Industry, ECGI Working Paper. 37

38 Listokin, Y. (2008). Management Always Wins the Close Ones, American Law and Economics Review, Vol. 10, Issue 2, pp Maug, E. and Rydqvist, K. (2009). Do Shareholders Vote Strategically? Voting Behavior, Proposal Screening, and Majority Rules, Review of Finance 13, pp McCrary, J. (2008). Manipulation of the running variable in the regression discontinuity design: A density test, Journal of Econometrics 142, pp Popadak, J. (2013). A Corporate Culture Channel: How Increased Shareholder Governance Reduces Firm Value, Working Paper, Wharton. Renneboog, L. and Szilagyi, P. (2011). The role of shareholder proposals in corporate governance, Journal of Corporate Finance 17, pp Romano, R. (2001). Less is more: Making institutional investor activism a valuable mechanism of corporate governance, Yale Journal on Regulation 18, pp Thomas, R. and Cotter, J. (2007). Shareholder proposals in the new millennium: Shareholder support, board response, and market reaction, Journal of Corporate Finance 13, pp U.S. House (2013). Examining the Market Power and Impact of Proxy Advisory Firms, Hearing before the Subcommittee on Capital Markets and Government Sponsored Enterprises of the Committee on Financial Services. Yermack, D. (2010). Shareholder Voting and Corporate Governance, Annual Review of Financial Economics 2, pp

39 9 Figures Figure 1: The Growth of Shareholder-sponsored Governance Proposals since 1997 This graph compares the number of S&P 1500 firms facing at least one majority-supported governance proposal with the number of proxy contests in S&P 1500 firms in a given year. The data for proxy contests has been kindly provided by Vyacheslav Fos using the methodology in in Fos (2013). The data for shareholder proposals comes from ISS. We define a proposal as majoritysupported if its support has reached more than 50% of votes cast for and against it Firms with majority-supported proposals Proxy contests

40 Figure 2: Thresholds and Samples 2.A: Thresholds / Voting Rules: F/(F+A) Official F/Outstanding CII Management ISS Figure 2.A illustrates and labels the different thresholds this paper focuses on. The CII threshold only considers votes for and against, while the ISS threshold counts abstentions, broker non-votes, and absent votes de facto as votes against the proposal, i.e., the approval threshold is 50% of votes for over shares outstanding. The official threshold, which we call the management threshold, is defined by the corporate bylaws. This threshold is fixed and known to shareholders before the shareholder meeting and can be based on any voting rule as discussed in Section 2.2., including the CII and ISS voting rule. In our main specifications, when evaluating the treatment effect of passing the CII (ISS) threshold, we require the CII (ISS) threshold to be different from the management threshold. 40

41 2.B: Specifications / Samples of interest: 2.B.1: Specification 1: Mngt-CII > 0.5% This specification requires that the number of votes counting against the proposal according to the CII and management rule differ by at least 0.5% of votes effectively cast for and against, i.e., there is a voting gap of at least 0.5%. As an example, consider the case when the management rule is to count abstentions in the denominator (which is the same thing as treating them as votes against the proposal) and the proposal obtains the following results: 501 for, 499 against and 50 abstentions. The voting result is 50.1% according to CII, 47.7% according to management, and the corresponding voting gap is 5% (i.e., 50/( )) of votes for and against. 2.B.2: Specification 2: Mngt-CII > 2% This specification requires that the number of votes counting against the proposal according to the CII and management rule differ by at least 2% of votes effectively cast for and against, i.e., there is a voting gap of at least 2%. 2.B.3: Specification 3: Full Sample CII Management ISS CII Management ISS The upper part Figure 2.B.3 shows the proposals in grey that are considered when evaluating ant treatment effect of passing the CII threshold (bold). The lower part shows the equivalent considered proposals in for an evaluation of the ISS threshold. In this specification all proposals are used. The treatment effects are estimated by running local linear regressions. 2.B.4: Specification 4: No Contamination Sample 41

42 CII Management ISS CII Management ISS The upper part Figure 2.B.4 shows the proposals in grey that are considered when evaluating ant treatment effect of passing the CII threshold (bold). To avoid contamination by passing also the management threshold, this specification focuses only on proposals that have not yet passed the management threshold. The lower part shows the equivalent considered proposals when evaluation of the ISS threshold. To avoid contamination by not passing the management threshold, this specification focuses only on proposals that have already passed the management threshold. The treatment effects are estimated by running local linear regressions. 2.B.5: Specification 5: Parametric estimation with controls CII-3% CII CII+3% ISS-3% ISS ISS+3% The upper part Figure 2.B.5 shows the proposals in grey that are considered when evaluating ant treatment effect of passing the CII threshold (bold). The lower part shows the equivalent considered proposals when evaluation of the ISS threshold. These specifications use only proposals that are contested, i.e., in which the voting outcome it in the range of 47-53% around the threshold of interest. This specification is estimated parametrically using OLS and controlling for potential effects of passing other thresholds as well. For example, we estimate the treatment effect of passing the CII threshold on outcome Y by running the following regression: Y i = α + β 1 Pass(CII) i + β 2 Pass(Mngt) i + β 3 Pass(ISS) i + ε i In this example the sample is restricted to proposals that reach support between 47 and 53% according to the CII voting rule; the coefficient of interest is β 1. 42

43 Figure 3: Distribution of Voting Shares Around the Approval Thresholds Proposals are grouped into one percentage-point bins: proposals that passed by between 0% and 1% are assigned to the first bin to the right of the red vertical line, and those that failed by similar margins are assigned to the first bin to the left of that line. The local linear regression is estimated using the bandwidth suggested by McCrary (2008). The first figure shows the results for proposals at the CII threshold (voting rule: For/(For+Against), the second figure at the Management threshold, and the third figure at the ISS threshold (voting rule: For/Outstanding). Proposals for which the threshold of interest overlaps (or differs by less than 0.5% of cast votes) with another threshold are excluded. Source : ISS ( ). 3. A: CII Threshold 43

44 3. B: ISS Threshold 44

45 Figure 4: Ex-post Implementation of Shareholder Proposals and Shareholder Voting Implementation is a dummy variable equal to one if the proposal is implemented in the year after the shareholder meeting in which a proposal is put to the vote. The first figure shows the results for proposals at the CII threshold (voting rule: For/(For+Against), the second figure at the Management threshold, and the third figure at the ISS threshold (voting rule: For/Outstanding). Proposals for which the threshold of interest overlaps (or differs by less than 0.5% of cast votes) with another threshold are excluded. The interval size of bin averages is chosen according to the methodology in Calonico et al. (2014). Source: DEF 14A filings ( ). 4. A: CII Threshold 45

46 4. B: Management Threshold 46

47 4. C: ISS Threshold 47

48 Figure 5: Future Voting against Directors and Shareholder Voting (Best Director) Votes against the best director are votes withheld corresponding to the nominee with the smallest amount of withholding among all management nominees at the meeting. The first set of figures shows the results for proposals at the CII threshold (voting rule: For/(For+Against) and the second set of figures at the ISS threshold (voting rule: For/Outstanding). Proposals for which the threshold of interest overlaps (or differs by less than 0.5% of cast votes) with another threshold are excluded. The interval size of bin averages is chosen according to the methodology in Calonico et al. (2014). Source: 10-Q, ISS ( ). 5. A: CII Threshold Voting against Best Director 48

49 49

50 5. B: ISS Threshold Voting against Best Director 50

51 Figure 6: Future Voting against Directors and Shareholder Voting (Worst Director) Votes against the worst director are votes withheld corresponding to the nominee with the largest amount of withholding among all management nominees at the meeting. The first set of figures shows the results for proposals at the CII threshold (voting rule: For/(For+Against) and the second set of figures at the ISS threshold (voting rule: For/Outstanding). Proposals for which the threshold of interest overlaps (or differs by less than 0.5% of cast votes) with another threshold are excluded. The interval size of bin averages is chosen according to the methodology in Calonico et al. (2014). Source: 10-Q, ISS ( ). 6. A: CII Threshold Voting against Worst Director 51

52 52

53 6.B: ISS Threshold Voting against Worst Director 53

54 Figure 7: Future ISS Recommendations against Directors and Shareholder Voting Recommendation against a director is a dummy variable that is equal to one of ISS recommends to withhold the vote for at least one management nominee. Recommendation against all directors is a dummy variable that is equal to one if ISS recommends to withhold the vote for all management nominees at the meeting. The figures show the results for proposals at the ISS threshold (voting rule: For/Outstanding). Proposals for which the threshold of interest overlaps (or differs by less than 0.5% of cast votes) with another threshold are excluded. The interval size of bin averages is chosen according to the methodology in Calonico et al. (2014). Source: 10-Q, ISS ( ). 7.A: ISS Threshold ISS Voting Recommendation against any Director 54

55 7.B: ISS Threshold ISS Voting Recommendation against all Directors 55

56 Figure 8: Passing a Governance Proposal Causes Bad Turnover Events Value-destroying turnovers are those that are announced less than two years after the shareholder meeting and lead to an announcement CAR [0;+1] below the 25 th percentile of the distribution. CARs are computed using the market model. The first figure shows the results for proposals at the CII threshold (voting rule: For/(For+Against), the second figure at the ISS threshold (voting rule: For/Outstanding). Proposals for which the threshold of interest overlaps (or differs by less than 0.5% of cast votes) with another threshold are excluded. The interval size of bin averages is chosen according to the methodology in Calonico et al. (2014). Source: ISS ( ). 8.A: CII Threshold 56